Reimagining Pricing Research with Generative AI: A Behind-the-Scenes Look at Our Work with Microsoft

by Mike Deinlein

“How much?”

It’s a starkly simple question and a notoriously difficult one to answer. Overly formulaic and non-representative pricing methodologies struggle to reflect how consumers and businesses make decisions in the real world. A Likert scale measuring purchase intent in a conjoint analysis or a Gabor-Granger price ladder does not represent a “decision” that people actually make. They either buy or they don’t buy.

In the consumer market, this process is usually instantaneous. I see a candy bar. It’s $1.50. I nearly always accept that, in the moment, to satisfy my overly gluttonous chocolate habit. But if you asked me in a survey… I have no idea what I’d say. I’m not thinking about it.

We’ll come back to the consumer example, but this “non-thinking” issue is even more pronounced when we consider a B2B environment. When we ask a purchase intent question, there is no real money at stake. A click of the button of “Definitely will buy” doesn’t represent the deliberate, analytical decision-making that accompanies large capital expenditures.

But what if, instead of relying on quick, click-based judgments, we engaged in a conversation? What if we encouraged participants to think through pricing more like they would in real business conversations? What if we replaced System 1 (fast, automatic) thinking with System 2 (slow, deliberate) thinking? Could we then have more confidence in our pricing elasticity curves?

A Classic Pricing Challenge Meets Cutting-Edge Tech

Our research partners at one of the largest, most well-respected technology companies had a problem. They didn’t trust the pricing curves they were getting with their Gabor-Granger methodology. You’ve probably used the Gabor-Granger pricing ladder – it’s simple, scalable, and provides easily digestible pricing curves. But the curves seemed overly inflated – suggesting a price tolerance much higher than what they saw among their B2B audience in the real world.

Their ask was simple – how could you reimagine pricing ladder research with AI? And more importantly, would it solve our curve-inflation problem? With Microsoft’s Azure-AI platform, we built a custom GPT chatbot that engaged developer managers—our client’s primary target audience—in a pricing negotiation, allowing respondents to answer in their natural language. The custom GPT understood the intent behind each response and surfaced either a higher or lower price in another conversational ask.

This more conversational interface slowed our respondents down and removed the quick, System-1 decision-making that we observed in the traditional Gabor-Granger. The Generative AI pricing moved the respondent task closer to the real world, where decisions are high stakes and involve System-2 thinking: slow, reasoned, and effortful. Comparing the results, the AI approach yielded more realistic and elastic demand curves. It reduced the risk of overpricing and protected against lower market penetration and revenue. In fact, market penetration was 58% higher and revenues 19% higher when relying on the AI-derived price versus the traditional Gabor-Granger results. An additional benefit is that we didn’t have to rely on a handful of pre-set prices. The AI tool had the flexibility to ask any price point across our price range, whereas the traditional Gabor-Granger was limited to the five pre-programmed prices. So instead of interpolating demand between price points, we had actual demand across every possible price.

What’s Next?

Our Generative AI pricing experiment proved that a purpose built custom-GPT is a viable and meaningful alternative to traditional pricing methodologies. However, it is just the beginning. We’ve already started refining the custom GPT to better reflect real-world negotiations. Pricing ladders are inherently artificial. We don’t ask a potential buyer if they’d purchase at $5, and if they say yes, we don’t then try to upsell them to $6. A more natural negotiation task where the respondent offers an opening price, Generative AI provides a counteroffer, and then we negotiate between those prices is a much more natural task and something we successfully built into our new custom GPT.

Secondly, we are actively experimenting with this methodology in the B2C space. We know Generative AI pricing works well for large capital expenditures in a B2B environment that requires slow, deliberate thinking. This same logic also applies to large B2C purchases that consumers mull over for longer than my impulsive chocolate purchases. However, even for smaller purchases, the conversational approach can be beneficial. A Gabor-Granger or a Van Westendorp are quick tasks, but they are still detached from the reality of in-the-moment purchase decisions. They aren’t measuring the same System 1 heuristics that culminate at the grocery shelf. Our Generative AI, conversational interface leverages System 2 principles, but it does a much better job at focusing the respondent on the product—an advantage with novel concept launches where pricing is less clearly defined and you want respondents to digest the benefits of a new product.

No matter the application, Generative AI pricing is redefining what’s possible in research. We are no longer constrained by what can be asked in a survey. Now we are only constrained by what can be asked in a conversation.

This work was originally presented by Microsoft at ESOMAR Congress 2024 held in Athens, Greece. The original white paper was submitted by Microsoft, showcasing the collaborative efforts and innovative insights achieved through our partnership.

Interested in reading more? Check out Mike’s other articles:

Aligning Brand and Customer Experience Strategies for Growth

Segmenting Key Drivers

As always, you can follow Burke, Inc. on our LinkedIn, Twitter, Facebook and Instagram pages.